1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * linux/include/linux/sunrpc/svc.h
4 *
5 * RPC server declarations.
6 *
7 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
8 */
9
10
11#ifndef SUNRPC_SVC_H
12#define SUNRPC_SVC_H
13
14#include <linux/in.h>
15#include <linux/in6.h>
16#include <linux/sunrpc/types.h>
17#include <linux/sunrpc/xdr.h>
18#include <linux/sunrpc/auth.h>
19#include <linux/sunrpc/svcauth.h>
20#include <linux/lwq.h>
21#include <linux/wait.h>
22#include <linux/mm.h>
23#include <linux/pagevec.h>
24
25/*
26 *
27 * RPC service thread pool.
28 *
29 * Pool of threads and temporary sockets. Generally there is only
30 * a single one of these per RPC service, but on NUMA machines those
31 * services that can benefit from it (i.e. nfs but not lockd) will
32 * have one pool per NUMA node. This optimisation reduces cross-
33 * node traffic on multi-node NUMA NFS servers.
34 */
35struct svc_pool {
36 unsigned int sp_id; /* pool id; also node id on NUMA */
37 struct lwq sp_xprts; /* pending transports */
38 atomic_t sp_nrthreads; /* # of threads in pool */
39 struct list_head sp_all_threads; /* all server threads */
40 struct llist_head sp_idle_threads; /* idle server threads */
41
42 /* statistics on pool operation */
43 struct percpu_counter sp_messages_arrived;
44 struct percpu_counter sp_sockets_queued;
45 struct percpu_counter sp_threads_woken;
46
47 unsigned long sp_flags;
48} ____cacheline_aligned_in_smp;
49
50/* bits for sp_flags */
51enum {
52 SP_TASK_PENDING, /* still work to do even if no xprt is queued */
53 SP_NEED_VICTIM, /* One thread needs to agree to exit */
54 SP_VICTIM_REMAINS, /* One thread needs to actually exit */
55};
56
57
58/*
59 * RPC service.
60 *
61 * An RPC service is a ``daemon,'' possibly multithreaded, which
62 * receives and processes incoming RPC messages.
63 * It has one or more transport sockets associated with it, and maintains
64 * a list of idle threads waiting for input.
65 *
66 * We currently do not support more than one RPC program per daemon.
67 */
68struct svc_serv {
69 struct svc_program * sv_program; /* RPC program */
70 struct svc_stat * sv_stats; /* RPC statistics */
71 spinlock_t sv_lock;
72 unsigned int sv_nrthreads; /* # of server threads */
73 unsigned int sv_maxconn; /* max connections allowed or
74 * '0' causing max to be based
75 * on number of threads. */
76
77 unsigned int sv_max_payload; /* datagram payload size */
78 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
79 unsigned int sv_xdrsize; /* XDR buffer size */
80 struct list_head sv_permsocks; /* all permanent sockets */
81 struct list_head sv_tempsocks; /* all temporary sockets */
82 int sv_tmpcnt; /* count of temporary sockets */
83 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
84
85 char * sv_name; /* service name */
86
87 unsigned int sv_nrpools; /* number of thread pools */
88 struct svc_pool * sv_pools; /* array of thread pools */
89 int (*sv_threadfn)(void *data);
90
91#if defined(CONFIG_SUNRPC_BACKCHANNEL)
92 struct lwq sv_cb_list; /* queue for callback requests
93 * that arrive over the same
94 * connection */
95 bool sv_bc_enabled; /* service uses backchannel */
96#endif /* CONFIG_SUNRPC_BACKCHANNEL */
97};
98
99/* This is used by pool_stats to find and lock an svc */
100struct svc_info {
101 struct svc_serv *serv;
102 struct mutex *mutex;
103};
104
105void svc_destroy(struct svc_serv **svcp);
106
107/*
108 * Maximum payload size supported by a kernel RPC server.
109 * This is use to determine the max number of pages nfsd is
110 * willing to return in a single READ operation.
111 *
112 * These happen to all be powers of 2, which is not strictly
113 * necessary but helps enforce the real limitation, which is
114 * that they should be multiples of PAGE_SIZE.
115 *
116 * For UDP transports, a block plus NFS,RPC, and UDP headers
117 * has to fit into the IP datagram limit of 64K. The largest
118 * feasible number for all known page sizes is probably 48K,
119 * but we choose 32K here. This is the same as the historical
120 * Linux limit; someone who cares more about NFS/UDP performance
121 * can test a larger number.
122 *
123 * For TCP transports we have more freedom. A size of 1MB is
124 * chosen to match the client limit. Other OSes are known to
125 * have larger limits, but those numbers are probably beyond
126 * the point of diminishing returns.
127 */
128#define RPCSVC_MAXPAYLOAD (1*1024*1024u)
129#define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
130#define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
131
132extern u32 svc_max_payload(const struct svc_rqst *rqstp);
133
134/*
135 * RPC Requests and replies are stored in one or more pages.
136 * We maintain an array of pages for each server thread.
137 * Requests are copied into these pages as they arrive. Remaining
138 * pages are available to write the reply into.
139 *
140 * Pages are sent using ->sendmsg with MSG_SPLICE_PAGES so each server thread
141 * needs to allocate more to replace those used in sending. To help keep track
142 * of these pages we have a receive list where all pages initialy live, and a
143 * send list where pages are moved to when there are to be part of a reply.
144 *
145 * We use xdr_buf for holding responses as it fits well with NFS
146 * read responses (that have a header, and some data pages, and possibly
147 * a tail) and means we can share some client side routines.
148 *
149 * The xdr_buf.head kvec always points to the first page in the rq_*pages
150 * list. The xdr_buf.pages pointer points to the second page on that
151 * list. xdr_buf.tail points to the end of the first page.
152 * This assumes that the non-page part of an rpc reply will fit
153 * in a page - NFSd ensures this. lockd also has no trouble.
154 *
155 * Each request/reply pair can have at most one "payload", plus two pages,
156 * one for the request, and one for the reply.
157 * We using ->sendfile to return read data, we might need one extra page
158 * if the request is not page-aligned. So add another '1'.
159 */
160#define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
161 + 2 + 1)
162
163/*
164 * The context of a single thread, including the request currently being
165 * processed.
166 */
167struct svc_rqst {
168 struct list_head rq_all; /* all threads list */
169 struct llist_node rq_idle; /* On the idle list */
170 struct rcu_head rq_rcu_head; /* for RCU deferred kfree */
171 struct svc_xprt * rq_xprt; /* transport ptr */
172
173 struct sockaddr_storage rq_addr; /* peer address */
174 size_t rq_addrlen;
175 struct sockaddr_storage rq_daddr; /* dest addr of request
176 * - reply from here */
177 size_t rq_daddrlen;
178
179 struct svc_serv * rq_server; /* RPC service definition */
180 struct svc_pool * rq_pool; /* thread pool */
181 const struct svc_procedure *rq_procinfo;/* procedure info */
182 struct auth_ops * rq_authop; /* authentication flavour */
183 struct svc_cred rq_cred; /* auth info */
184 void * rq_xprt_ctxt; /* transport specific context ptr */
185 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
186
187 struct xdr_buf rq_arg;
188 struct xdr_stream rq_arg_stream;
189 struct xdr_stream rq_res_stream;
190 struct page *rq_scratch_page;
191 struct xdr_buf rq_res;
192 struct page *rq_pages[RPCSVC_MAXPAGES + 1];
193 struct page * *rq_respages; /* points into rq_pages */
194 struct page * *rq_next_page; /* next reply page to use */
195 struct page * *rq_page_end; /* one past the last page */
196
197 struct folio_batch rq_fbatch;
198 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
199 struct bio_vec rq_bvec[RPCSVC_MAXPAGES];
200
201 __be32 rq_xid; /* transmission id */
202 u32 rq_prog; /* program number */
203 u32 rq_vers; /* program version */
204 u32 rq_proc; /* procedure number */
205 u32 rq_prot; /* IP protocol */
206 int rq_cachetype; /* catering to nfsd */
207 unsigned long rq_flags; /* flags field */
208 ktime_t rq_qtime; /* enqueue time */
209
210 void * rq_argp; /* decoded arguments */
211 void * rq_resp; /* xdr'd results */
212 __be32 *rq_accept_statp;
213 void * rq_auth_data; /* flavor-specific data */
214 __be32 rq_auth_stat; /* authentication status */
215 int rq_auth_slack; /* extra space xdr code
216 * should leave in head
217 * for krb5i, krb5p.
218 */
219 int rq_reserved; /* space on socket outq
220 * reserved for this request
221 */
222 ktime_t rq_stime; /* start time */
223
224 struct cache_req rq_chandle; /* handle passed to caches for
225 * request delaying
226 */
227 /* Catering to nfsd */
228 struct auth_domain * rq_client; /* RPC peer info */
229 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
230 struct task_struct *rq_task; /* service thread */
231 struct net *rq_bc_net; /* pointer to backchannel's
232 * net namespace
233 */
234 unsigned long bc_to_initval;
235 unsigned int bc_to_retries;
236 void ** rq_lease_breaker; /* The v4 client breaking a lease */
237 unsigned int rq_status_counter; /* RPC processing counter */
238};
239
240/* bits for rq_flags */
241enum {
242 RQ_SECURE, /* secure port */
243 RQ_LOCAL, /* local request */
244 RQ_USEDEFERRAL, /* use deferral */
245 RQ_DROPME, /* drop current reply */
246 RQ_VICTIM, /* Have agreed to shut down */
247 RQ_DATA, /* request has data */
248};
249
250#define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net)
251
252/*
253 * Rigorous type checking on sockaddr type conversions
254 */
255static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
256{
257 return (struct sockaddr_in *) &rqst->rq_addr;
258}
259
260static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
261{
262 return (struct sockaddr_in6 *) &rqst->rq_addr;
263}
264
265static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
266{
267 return (struct sockaddr *) &rqst->rq_addr;
268}
269
270static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
271{
272 return (struct sockaddr_in *) &rqst->rq_daddr;
273}
274
275static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
276{
277 return (struct sockaddr_in6 *) &rqst->rq_daddr;
278}
279
280static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
281{
282 return (struct sockaddr *) &rqst->rq_daddr;
283}
284
285/**
286 * svc_thread_should_stop - check if this thread should stop
287 * @rqstp: the thread that might need to stop
288 *
289 * To stop an svc thread, the pool flags SP_NEED_VICTIM and SP_VICTIM_REMAINS
290 * are set. The first thread which sees SP_NEED_VICTIM clears it, becoming
291 * the victim using this function. It should then promptly call
292 * svc_exit_thread() to complete the process, clearing SP_VICTIM_REMAINS
293 * so the task waiting for a thread to exit can wake and continue.
294 *
295 * Return values:
296 * %true: caller should invoke svc_exit_thread()
297 * %false: caller should do nothing
298 */
299static inline bool svc_thread_should_stop(struct svc_rqst *rqstp)
300{
301 if (test_and_clear_bit(nr: SP_NEED_VICTIM, addr: &rqstp->rq_pool->sp_flags))
302 set_bit(nr: RQ_VICTIM, addr: &rqstp->rq_flags);
303
304 return test_bit(RQ_VICTIM, &rqstp->rq_flags);
305}
306
307struct svc_deferred_req {
308 u32 prot; /* protocol (UDP or TCP) */
309 struct svc_xprt *xprt;
310 struct sockaddr_storage addr; /* where reply must go */
311 size_t addrlen;
312 struct sockaddr_storage daddr; /* where reply must come from */
313 size_t daddrlen;
314 void *xprt_ctxt;
315 struct cache_deferred_req handle;
316 int argslen;
317 __be32 args[];
318};
319
320struct svc_process_info {
321 union {
322 int (*dispatch)(struct svc_rqst *rqstp);
323 struct {
324 unsigned int lovers;
325 unsigned int hivers;
326 } mismatch;
327 };
328};
329
330/*
331 * List of RPC programs on the same transport endpoint
332 */
333struct svc_program {
334 struct svc_program * pg_next; /* other programs (same xprt) */
335 u32 pg_prog; /* program number */
336 unsigned int pg_lovers; /* lowest version */
337 unsigned int pg_hivers; /* highest version */
338 unsigned int pg_nvers; /* number of versions */
339 const struct svc_version **pg_vers; /* version array */
340 char * pg_name; /* service name */
341 char * pg_class; /* class name: services sharing authentication */
342 enum svc_auth_status (*pg_authenticate)(struct svc_rqst *rqstp);
343 __be32 (*pg_init_request)(struct svc_rqst *,
344 const struct svc_program *,
345 struct svc_process_info *);
346 int (*pg_rpcbind_set)(struct net *net,
347 const struct svc_program *,
348 u32 version, int family,
349 unsigned short proto,
350 unsigned short port);
351};
352
353/*
354 * RPC program version
355 */
356struct svc_version {
357 u32 vs_vers; /* version number */
358 u32 vs_nproc; /* number of procedures */
359 const struct svc_procedure *vs_proc; /* per-procedure info */
360 unsigned long __percpu *vs_count; /* call counts */
361 u32 vs_xdrsize; /* xdrsize needed for this version */
362
363 /* Don't register with rpcbind */
364 bool vs_hidden;
365
366 /* Don't care if the rpcbind registration fails */
367 bool vs_rpcb_optnl;
368
369 /* Need xprt with congestion control */
370 bool vs_need_cong_ctrl;
371
372 /* Dispatch function */
373 int (*vs_dispatch)(struct svc_rqst *rqstp);
374};
375
376/*
377 * RPC procedure info
378 */
379struct svc_procedure {
380 /* process the request: */
381 __be32 (*pc_func)(struct svc_rqst *);
382 /* XDR decode args: */
383 bool (*pc_decode)(struct svc_rqst *rqstp,
384 struct xdr_stream *xdr);
385 /* XDR encode result: */
386 bool (*pc_encode)(struct svc_rqst *rqstp,
387 struct xdr_stream *xdr);
388 /* XDR free result: */
389 void (*pc_release)(struct svc_rqst *);
390 unsigned int pc_argsize; /* argument struct size */
391 unsigned int pc_argzero; /* how much of argument to clear */
392 unsigned int pc_ressize; /* result struct size */
393 unsigned int pc_cachetype; /* cache info (NFS) */
394 unsigned int pc_xdrressize; /* maximum size of XDR reply */
395 const char * pc_name; /* for display */
396};
397
398/*
399 * Function prototypes.
400 */
401int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
402void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
403int svc_bind(struct svc_serv *serv, struct net *net);
404struct svc_serv *svc_create(struct svc_program *, unsigned int,
405 int (*threadfn)(void *data));
406struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
407 struct svc_pool *pool, int node);
408bool svc_rqst_replace_page(struct svc_rqst *rqstp,
409 struct page *page);
410void svc_rqst_release_pages(struct svc_rqst *rqstp);
411void svc_rqst_free(struct svc_rqst *);
412void svc_exit_thread(struct svc_rqst *);
413struct svc_serv * svc_create_pooled(struct svc_program *prog,
414 struct svc_stat *stats,
415 unsigned int bufsize,
416 int (*threadfn)(void *data));
417int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
418int svc_pool_stats_open(struct svc_info *si, struct file *file);
419void svc_process(struct svc_rqst *rqstp);
420void svc_process_bc(struct rpc_rqst *req, struct svc_rqst *rqstp);
421int svc_register(const struct svc_serv *, struct net *, const int,
422 const unsigned short, const unsigned short);
423
424void svc_wake_up(struct svc_serv *);
425void svc_reserve(struct svc_rqst *rqstp, int space);
426void svc_pool_wake_idle_thread(struct svc_pool *pool);
427struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv);
428char * svc_print_addr(struct svc_rqst *, char *, size_t);
429const char * svc_proc_name(const struct svc_rqst *rqstp);
430int svc_encode_result_payload(struct svc_rqst *rqstp,
431 unsigned int offset,
432 unsigned int length);
433unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
434 struct xdr_buf *payload);
435char *svc_fill_symlink_pathname(struct svc_rqst *rqstp,
436 struct kvec *first, void *p,
437 size_t total);
438__be32 svc_generic_init_request(struct svc_rqst *rqstp,
439 const struct svc_program *progp,
440 struct svc_process_info *procinfo);
441int svc_generic_rpcbind_set(struct net *net,
442 const struct svc_program *progp,
443 u32 version, int family,
444 unsigned short proto,
445 unsigned short port);
446int svc_rpcbind_set_version(struct net *net,
447 const struct svc_program *progp,
448 u32 version, int family,
449 unsigned short proto,
450 unsigned short port);
451
452#define RPC_MAX_ADDRBUFLEN (63U)
453
454/*
455 * When we want to reduce the size of the reserved space in the response
456 * buffer, we need to take into account the size of any checksum data that
457 * may be at the end of the packet. This is difficult to determine exactly
458 * for all cases without actually generating the checksum, so we just use a
459 * static value.
460 */
461static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
462{
463 svc_reserve(rqstp, space: space + rqstp->rq_auth_slack);
464}
465
466/**
467 * svcxdr_init_decode - Prepare an xdr_stream for Call decoding
468 * @rqstp: controlling server RPC transaction context
469 *
470 */
471static inline void svcxdr_init_decode(struct svc_rqst *rqstp)
472{
473 struct xdr_stream *xdr = &rqstp->rq_arg_stream;
474 struct xdr_buf *buf = &rqstp->rq_arg;
475 struct kvec *argv = buf->head;
476
477 WARN_ON(buf->len != buf->head->iov_len + buf->page_len + buf->tail->iov_len);
478 buf->len = buf->head->iov_len + buf->page_len + buf->tail->iov_len;
479
480 xdr_init_decode(xdr, buf, p: argv->iov_base, NULL);
481 xdr_set_scratch_page(xdr, page: rqstp->rq_scratch_page);
482}
483
484/**
485 * svcxdr_init_encode - Prepare an xdr_stream for svc Reply encoding
486 * @rqstp: controlling server RPC transaction context
487 *
488 */
489static inline void svcxdr_init_encode(struct svc_rqst *rqstp)
490{
491 struct xdr_stream *xdr = &rqstp->rq_res_stream;
492 struct xdr_buf *buf = &rqstp->rq_res;
493 struct kvec *resv = buf->head;
494
495 xdr_reset_scratch_buffer(xdr);
496
497 xdr->buf = buf;
498 xdr->iov = resv;
499 xdr->p = resv->iov_base + resv->iov_len;
500 xdr->end = resv->iov_base + PAGE_SIZE;
501 buf->len = resv->iov_len;
502 xdr->page_ptr = buf->pages - 1;
503 buf->buflen = PAGE_SIZE * (rqstp->rq_page_end - buf->pages);
504 xdr->rqst = NULL;
505}
506
507/**
508 * svcxdr_encode_opaque_pages - Insert pages into an xdr_stream
509 * @xdr: xdr_stream to be updated
510 * @pages: array of pages to insert
511 * @base: starting offset of first data byte in @pages
512 * @len: number of data bytes in @pages to insert
513 *
514 * After the @pages are added, the tail iovec is instantiated pointing
515 * to end of the head buffer, and the stream is set up to encode
516 * subsequent items into the tail.
517 */
518static inline void svcxdr_encode_opaque_pages(struct svc_rqst *rqstp,
519 struct xdr_stream *xdr,
520 struct page **pages,
521 unsigned int base,
522 unsigned int len)
523{
524 xdr_write_pages(xdr, pages, base, len);
525 xdr->page_ptr = rqstp->rq_next_page - 1;
526}
527
528/**
529 * svcxdr_set_auth_slack -
530 * @rqstp: RPC transaction
531 * @slack: buffer space to reserve for the transaction's security flavor
532 *
533 * Set the request's slack space requirement, and set aside that much
534 * space in the rqstp's rq_res.head for use when the auth wraps the Reply.
535 */
536static inline void svcxdr_set_auth_slack(struct svc_rqst *rqstp, int slack)
537{
538 struct xdr_stream *xdr = &rqstp->rq_res_stream;
539 struct xdr_buf *buf = &rqstp->rq_res;
540 struct kvec *resv = buf->head;
541
542 rqstp->rq_auth_slack = slack;
543
544 xdr->end -= XDR_QUADLEN(slack);
545 buf->buflen -= rqstp->rq_auth_slack;
546
547 WARN_ON(xdr->iov != resv);
548 WARN_ON(xdr->p > xdr->end);
549}
550
551/**
552 * svcxdr_set_accept_stat - Reserve space for the accept_stat field
553 * @rqstp: RPC transaction context
554 *
555 * Return values:
556 * %true: Success
557 * %false: No response buffer space was available
558 */
559static inline bool svcxdr_set_accept_stat(struct svc_rqst *rqstp)
560{
561 struct xdr_stream *xdr = &rqstp->rq_res_stream;
562
563 rqstp->rq_accept_statp = xdr_reserve_space(xdr, XDR_UNIT);
564 if (unlikely(!rqstp->rq_accept_statp))
565 return false;
566 *rqstp->rq_accept_statp = rpc_success;
567 return true;
568}
569
570#endif /* SUNRPC_SVC_H */
571

source code of linux/include/linux/sunrpc/svc.h